AWS in Orbit: Leveraging generative AI to do more at the rugged space edge with AWS.

[ Music ]

>> Maria Varmazis: Welcome to "AWS in Orbit." We've collaborated with AWS on a fantastic project at the intersection of space technology, cloud computing, and artificial intelligence. I'm Maria Varmazis. We'll explore not just what's possible but what's meaningful in the realm of space and cloud innovation. On "AWS in Orbit," we grapple with the complex challenges and unparalleled opportunities that arise when we use space to address pressing issues right here on Earth. From enhancing global security to monitoring climate change, to democratizing education. The implications are as limitless as space itself.

[ Music ]

Episode one, Kathy O'Donnell, AWS and generative AI.

>> Kathy O'Donnell: My name is Kathy O'Donnell and --

>> Maria Varmazis: Kathy O'Donnell is the leader of Space Solutions Architecture for AWS.

>> Kathy O'Donnell: We're a part of aerospace and satellite at AWS, so we focus specifically on aerospace and satellite customers.

>> Maria Varmazis: In this extended conversation, we dive into how AWS is supporting generative AI in the space domain. She walks us through some incredible case studies with AWS customers like D-Orbit and Latitude 040 who are using generative AI and space technologies to improve life here on Earth. Can you tell me a little bit about how you got into the role that you're in now at AWS?

>> Kathy O'Donnell: I started off as an applied research mathematician with the Department of Defense. So I started back in 2000 and got to be involved in the transition from, like landlines and telephones to the internet and cellular networks and the use of satellite communications. And at the same time, because I was a mathematician, I did a lot of modeling and analytics. And it was great because I had access to like super computers and all sorts of data.

After it was about 14 years, I had just worked with the government, and I was sort of like, I want to try something new. I want to see what else is out there. Like I'm sort of seeing another 30 years of career ahead of me. And I'm like, you know, I like to try new things. So I actually went to be a director of data science at Capital One's Innovation Lab in San Francisco. And it was a great fit because that job was applying machine learning and new techniques to kind of classic problems in credit cards and finance. And one of the areas I found that I was really interested in was fraud and money laundering.

After Capital One I actually cofounded and led technical development in a startup searching for evidence of money laundering and fraud using analytics and machine learning models. Because I didn't have any resources, we had no money for anything, I basically had --

>> Maria Varmazis: Set of story.

>> Kathy O'Donnell: Yeah, I had a Mac mini on my desk, which was very different from my two previous jobs. I was like alright, I'm using AWS because, you know, for very reasonable cost I can get access to a lot of compute resources. I can get access to, you know, creating a platform on a website. So that's where I really got my knowledge of AWS from was building my own platform.

And when I was ready to transition from there, a friend of mine pointed me to a new job opening at AWS and said, hey, there's a new team called Aerospace and Satellite. They've only been around for about a year. AWS has decided to really focus on aerospace and satellite customers. We want to create this team to very specifically focus on bringing the cloud to space.

>> Maria Varmazis: Did you know what that meant before you applied for the job, like bringing the cloud to space? Like, was that a phrase that meant something to you, or was that kind Of like what is that?

>> Kelly O'Donnell: So it did. So, well I saw the posting and then I started researching because I've been kicking around in finance for a while, so I hadn't really been paying attention to space. I started looking into what companies were doing with space technology, and a lot of times all you think about is something like providing internet service or GPS. The goal here was something that I thought was great, that we're transforming the future of space. We're going to develop some innovative solutions.

One that had recently come out was called Ground Station. So it's a pay as you go service to download data from your satellites directly into AWS. And I'm a huge fan of pay as you go, especially in IT infrastructure because technology moves so quickly that it can be out of date like before you even know it. And you have to keep a whole team involved to keep it up to date. You've got to spend more money. So yeah, to me it made a lot of sense, especially for space companies.

In addition to, what, what really interested me was the use cases. So I went and checked out everything that the aerospace and satellite group had posted on AWS and particularly some of the customer use cases. There's a lot going on, especially in terms of like Earth observation and doing things like sustainability work, looking at environmental impacts. And something that was really personally important to me because I was in Northern Nevada. I had grown up in Reno. Forest fires are a constant issue. And, and a lot of people who live in Northern California and that area know. There was a company called XE [phonetic], and they're an Australian company. They were able to take, and I believe the numbers are two and a half million images every day from ground cameras plus 30 gigs of satellite imagery. Put that together to immediately detect the beginning of forest fires or brush fires. And within minutes let first responders know that something is popping off.

And that to me was so amazing, that that concept of data fusion and speed. The whole reason that we collect all of this data is to take action, right, to move it to insight and move it to action. By using AWS, that amount of storage with the analytics that we provide within minutes so as soon as you start seeing those plumes of smoke to say, hey, go check this out. Something's going on there. I mean that jus- that saves the first responders. These are very dangerous situations when the fire- forest fires get out of control. It saves the people living there. For me, that was very inspirational and really made me start thinking about how we can use both cloud and space to- to help change the way that we live on Earth and make it better.

>> Maria Varmazis: Yeah, it made it really real for you when sometimes when we talk about cloud and space, it seems very abstract and kind of like, well, who is that for? What is that going to do? But that's a real life and death, life improvement situation, which is incredibly powerful. I can totally see how that would speak to you.

>> Kathy O'Donnell: Yeah, I mean, I always appreciated like the things that our space agencies do, and it's always very exciting to see, you know, space shuttles go and what's happening on the ISS. But yeah, to me that really brought it home. That, you know, this- the things we do in space are not just for research or for, you know, for decades down the road. They can help right now.

And by giving people, like anyone, including like companies in Australia access to compute, access to cloud storage access to space data. Like they could come up with the most creative and interesting ideas to change the way that we live in positive ways, to make that positive impact on the way we live.

[ Music ]

>> Maria Varmazis: That, and that's so interesting that a company in Australia is saving lives in America. That's just, that, you know, other side of the planet and still really making a real life impact.

>> Kathy O'Donnell: Yeah, and so what really appealed to me was that we could really stretch our capabilities across the planet to help people who might be in dire situations. But at the same time, get access to the internet for people who don't usually have access, that the fiber lines haven't been laid out to, who are in remote locations. Give them the opportunity to access data and do super cool things with it. Because, you know, this idea of using satellite imagery to immediately send first responders, that's a new and innovative capability coming out of a startup.

You don't necessarily need all of these- all of this money and all of these resources to have a great idea. And so what's really appealing to me about AWS in particular and other cloud companies is that you can have these amazing ideas and test them. And now you're able to test these with space data, which traditionally has not been available to everyone. Like you had to work for big research institute or a governmental agency for a huge corporation. But now this data is there, and you can access it and you can, you know, do whatever crazy, inventive, innovative thing that you want to do.

And so I really love how we can provide these resources to everyone in the world because, I don't- I don't necessarily know, right what everyone needs but, but being able to give that capability to people so that they can come up with those ideas and support like the development of them, I think that's really what- what drew me in.

>> Maria Varmazis: I- I can hear that in your voice. You really personally connected with that. What a great job to have where you feel like that mission, right. Like, I'm, like, kind of geeking out with you. Like that's really cool.

>> Kathy O'Donnell: I just love it. Like people who have never had access to the internet, uou know, they know what they need, but they in the past never had the resources to do it. Now it's like, yep, you can do it. Just, you know, five bucks you can do.

>> Maria Varmazis: Well how does cloud come into play when we're talking about these incredible things that space can do for folks on the ground? And how does AI also work into that?

>> Kathy O'Donnell: So I think it comes in in a, in a few different ways. The first is just dealing with the massive amounts of data. So you can imagine if you're taking imagery of the Earth. That is petabytes of data coming in. So being able to store that on a cloud rather than having to spend the money to have your own storage for that really helps our customers. So they can focus more on the mission, what they want to do with that imagery, rather than having to deal with storing and all of that imagery.

And that's also where AI comes into play as well. Because once you have that stored, you want to do something with it. And so when you have it stored in AWS, if you build your analysis, your analytics, your machine learning on AWS, you can just immediately put it against the data that you have stored. And so you get almost near real time or real time results.

So for a company that's doing something like wildfire detection, very important that, you know, data comes off of satellite. Goes into storage, gets immediately put through analysis. And then gets sent out to people who can do something about it and actually take action.

>> Maria Varmazis: And, and that's not a small revolution in how things are done. I mean, not that long ago, I mean even mow still, would you like to walk us through a little bit about the, and it's sort of a weird question. But that's not how it's always done necessarily. In some cases, that data takes forever to get where it needs to go, literally physically mailing somebody, maybe a hard drive or something.

>> Kathy O'Donnell: Yeah, like, like in the past, and, and you know, back in my day, you know, it was stored on cassette tape. It was stored on disk, and you had to, you know, it would come down. Get stored by some organization over here. You'd have to physically go send, in my case, an intern to go pick it up or send it through the mail. So there's a few hours of delay. Hours at a minimum but more days, weeks. Then you have to load it up into your systems.

You have to run, you know, probably some pre-filtering or some screening on the data because satellite imagery, there's a lot of stuff that isn't too interesting in there. A lot of, you know, pictures of the ocean or clouds that, you know, you need to filter through until you get to things that you actually can take action on. So that takes time. And yeah, and then actually publishing those results. So then what do you do? You have to call someone or, you know, you put it in a weekly report. But now because we have our cloud capabilities along with some of these space capabilities, yeah, we're talking milliseconds until we can actually like take action, so.

>> Maria Varmazis: Right. So there's being able to do that incredible heavy computing lift quickly, and then there's doing all that even in a challenging environment like space, the ultimate edge computing really. So what does that look like?

>> Kathy O'Donnell: So, so this is an experiment that we've been running real world experiment that we launched in 2022. We worked with D-Orbit who provided the satellite and Unibap who provided the space computer. We actually put AWS software on the satellite so that on the ground terrestrially, the idea is basically you can sign into your AWS account, build a machine learning model, build an analytic and just send it to the satellite. Anyone can do this.

[ Music ]

>> Maria Varmazis: That's another cool thing about some of the capabilities that I'm going to be talking about. You don't have to be a space expert. You don't have to be a mathematician or an engineer, right. In fact, in this experiment we had, we did have data science, scientists building the models, but they weren't space focused data scientists. They had not done space stuff.

We had them build a few different models, one in particular detected clouds in imagery. So they built that using our tools. We put it on the satellite, and it was able to reduce the size of the images that we needed to send back down by 42%.

>> Maria Varmazis: Wow.

>> Kathy O'Donnell: Which is great.

>> Maria Varmazis: It's huge.

>> Kathy O'Donnel: Yeah. So without space knowledge with just, you know, some, some data science background, they were able to build that model. And with the experiment we ran, we were able to just push that model unmodified to the satellite so that it could run on the imagery.

>> Maria Varmazis: That's incredible. That's an amazing efficiency. After that was deployed, I mean I imagine all sorts of great things that were seeing afterwards from the team on the ground. Like was there any outcome afterwards that you all observed that you all observed?

>> Kathy O'Donnell: Yeah. Well, so for, for our experiments, we, we tried to define them very narrowly because this was our first experiment. You got to be careful when you're doing space stuff. Because yeah -

>> Maria Varmazis: I, I understand that.

>> Kathy O'Donnell: But -

>> Maria Varmazis: Totally understood. Mm-hmm.

>> Kathy O'Donnell: But yeah, just the fact that we were able to do that, we did, we also had anomaly detection running up on this satellite. We were able to show that we could do principal component analysis on imagery.

So we were able to do some really interesting things on the satellite, and it ran exactly the way that we expected it to. Which is, you know -

>> Maria Varmazis: That's great.

>> Kathy O'Donnell: in software world, that's amazing. Like yay, it ran. But I think that the cloud detection is pretty big because and, and that has some big impacts as well. Like we don't take a lot of imagery of the ocean or of clouds, right, because we don't want to clog up a restricted down lake to, to our ground stations. But that has some problems, like if there is an incident in the ocean. If there is a volcano happening. If you know, there's something that, that's happening in a place we don't normally take imagery of, then we don't have a record of it. But if we're able to run up on the satellite, a model saying, oh, hey, this this is actually interesting. This section of the ocean is interesting right now. Let's take pictures. Then --

>> Maria Varmazis: And you can tell hey, edge cases, right?

>> Kathy O'Donnell: Yeah, yeah, keep the edge cases and send those back down.

>> Maria Varmazis: Absolutely.

>> Kathy O'Donnell: Because otherwise it's, it's- right now it's all or nothing on that imagery. So and we have because, you know, the laws of physics, have to default to nothing, so.

>> Maria Varmazis: Those pesky laws of physics.

>> Kathy O'Donnell: Yeah. So our traditional machine learning is really exciting, and we're working on that. But we are working on generative AI, and this has been, I mean, I was already excited about my job. But then when they were like, hey, we really want to start working on generative AI. There's been some tremendous improvements in like transformer architectures. Really weird things are going on with generative AI. Can we apply that to space? I like raise my hand. I'm like, yes, I'm in Let me see what I can do to, to help our customers.

Because in this too to me kind of, this is another thing where giving people the opportunity to use it is so important. Because generative AI, the, the potential there is huge. There's a lot of like technological flash in the pan type things. But generative AI is, is something that I think will really change the way that we do things. And so being able or being a part of making that available to people everywhere was really exciting. Because right now you can again just sign on to AWS and have access to large language models.

So like Claude two, you can feed Claude 2 a book and start asking it questions. And it's so different. Like I'm kind of an old school machine learning person, so asking questions of a machine still feels really weird to me, and I'm trying to get used to prompt engineering. But, but yeah, but in space we found quite a few really interesting use cases.

I think a lot of people think about generative AI, and they're like, oh, it's just making weird pictures or it's helping people cheat on tests or essays. But we were looking at it from the, the side of how can we help our customers? What do they do that is difficult that generative AI could really provide some assistance.

>> Maria Varmazis: So yeah, speaking of use cases, there's one that's really interesting that humanizes the technology, and they're called Latitude 040. So okay, can you tell me a little bit about who they are and what they do?

>> Kathy O'Donnell: Yeah. So Latitude 040 is an Italian company, and they've been working with the Torino City Lab and using AWS to create an application that provides indicators of urban sustainability. So they process huge amounts of data from satellite images, from surveys, from ground cameras, just to help, you know, the city of Torino in particular for this application, determine how to make their city live better for their residents.

This first application, they're actually bringing down lots of data from satellite imagery so that you can see kind of the, the lay of the land, like how things look from above. But what's really cool with Latitude 040 since we've introduced these generative AI capabilities into AWS, they're able to go even further. So they're able to work with urban planners to produce what we call digital twin simulations.

Basically, you have your real city, but then you have your digital twin of the city, which you can start doing experiments with. Things that you couldn't do with the real city itself. Like you can't just, you know, rip out a street and put in a sidewalk in the real city. That's, that's a lot of money. In a digital twin, though, you can do those experiments. And then you run it through simulations and testing to see how that impacts your digital twin.

And with generative AI, there's a couple of ways that you can go with this. So you can ask, you know, if I make this change, what's the impact? So if I replace this four lane road with a bike path and park land, what's the impact to the city? Do I bring down the heat index? Do I actually make traffic worse in other places because I've removed this road? How does this impact my city holistically?

The other way that you can use generative AI is to say, okay, we've kind of exhausted the ideas we have for changing things. What are other things we may be able to change within these constraints? And generative AI can very quickly and fairly inexpensively put out a lot of new ideas about what you can change in your digital twin subject to kind of the constraints that you've given it, that you can then test.

They're able to create this digital twin and then identify which areas of a city are the most suitable for like nature based applications. And then see how that works for the city itself at a much lower cost than we had before. So you don't- you don't have to actually go do it. You can test it out on your digital twin, and then the one that works out the best, the one that, you know, has the most impact to people that optimizes nature and the way that people live within your city. You can then say okay, let's, let's do that in real life. Let's actually go replace this road.

>> Maria Varmazis: That's amazing. And especially now, as many, many cities that are talking about climate resiliency and trying to make sure that, you know, people are, are able to, you know, deal with the incredible heat in the summer that's happening now. That just feels like such a- there's so much urgency behind this, but it's so important. And to be able to make that kind of a plan is, it's, it's really incredible. I mean, that's sounds like almost sci-fi. But it's amazing that we can do.

>> Kathy O'Donnell: Yeah, and I think what's, what's super cool about it is, yeah, if you have a room full of experts, you can come up with quite a few scenarios, right. They're going to be very good with generative AI, though you can get as many scenarios as you want to test. And you know- and then have your experts look at those. So it really reduces the amount of time. It allows you to experiment a lot more. And also. generative AI has crazy ideas. So you know, some of them are terrible, but other ones are like, oh, wow, like we maybe wouldn't have thought about it that way because we're humans and we think, you know, you should always have a sidewalk there. And you should always have a bridge there. But generative AI is coming in and being like, nope, I've, you know, let's try something a little wacky.

>> Maria Varmazis: I wanted to stay in Italy for a minute and talk about D-Orbit next as well, because I, I'd love to hear the story about how they're working with Unibap. So two European companies doing something really cool. Can you tell me about it?

>> Kathy O'Donnell: Yeah. So we have a real world experiment running right now. My, my team worked very closely with D-Orbit and Unibap on this. So D-Orbit provided the satellite in early 2022. Unibap provided the space computer, and at AWS we provided the software. AWS IOT Greengrass helps people control edge devices, so their internet of things edge devices. And by using it with satellites, we kind of transform the way we think about satellites and start thinking about them as edge components. So you know of edge stuff in your own life. That's your cellphone, that's your ring doorbell, that's your Alexa.

So now a satellite is one of those things. So we actually put AWS Greengrass on the D-Orbit satellite and the Unibap space computer, along with some extra networking capability. So as, as you and your listeners may know, satellites communicate with the ground, generally through RF. This, this can be hard for people because not a lot of things communicate over RF anymore. We're used to using the internet and using packet based communications. So with this networking capability and the software on the satellite, we're able to actually develop software in AWS and just push it to the satellite.

So you don't need to be an expert in satellite communication technology. You don't need to be an expert in like the intricacies of space computing. And as part of this experiment, we actually had some of our data scientists build models in Amazon SageMaker, packaged them up and put them on the satellite. Amazon SageMaker is our machine learning platform. And on Amazon SageMaker, you can build models. You can use other people's models. You can bring in a lot of data, create analytics. So it just makes the model building life cycle a lot easier for data scientists and machine learning engineers.

We have a tool called SageMaker Jumpstart where you're able to literally three clicks deploy a foundation model and start creating images or start chatting to a chat bot. I'm an old school tech person at this point, and I still kind of don't believe that I was able to do it so quickly. That, that I was able to deploy like a stable diffusion image generation model and then just start asking it for things. I want to see a puppy on a Unicorn. I want to see a lunar habitat. And that again just backs up this idea of making it available to anyone.

And these data scientists art space people, they are data scientists. And, you know, we gave them some examples of models to build. These are very classic use cases for people running Earth observation workloads, so people who want to do things with images of the Earth. One of the main ones there was cloud detection. So as you can imagine, as you're flying over the Earth taking lots of images, a lot of it is clouds. And I love clouds, but --

>> Maria Varmazis: They're kind of annoying sometimes.

>> Kathy O'Donnell: They kind of look the same. There's not a lot of information to be extracted from many pictures of clouds. So they developed cloud masking algorithm. So on the satellite itself, it's able to look at the pictures, mask out the cloud and reduce the image sizes. And in testing, we saw the reduction in image sizes, image size be up to 42%. And given the limitations of that downlink from a satellite to the ground, 42% reduction means that we can send down a lot more interesting pictures.

>> Maria Varmazis: Yeah, and not have everyone have to comb through cloud pictures and trying to figure out what's actually important, yeah.

>> Kathy O'Donnell: You could use this in very important and life threatening situations. So for instance, as we fly over the oceans, we don't take a lot of pictures. The ocean is just the ocean most of the time and covers most of the Earth and is water. A lot of cool stuff going on underneath, but pictures of it not, not a lot. But sometimes very interesting things do happen in the ocean, right.

We have airplanes flying over. We have ships in the ocean. You potentially have volcanoes. You have earthquakes that create tsunamis. And so if you have this capability on orbit to say, hey, something interesting is happening and that can send an alert down to the ground, that just speeds up our time to action. That allows us to start taking more pictures so that we know what's going on. Allows us to get people out there in case there's an emergency or get people evacuated from somewhere that might be hit.

So the faster we can do that, the more lives we can save. This is sort of that first step, right, of alright, we can do it. We can do on orbit compute, and we can do it with, with software that's been created by people who know about software and know about the mission that they want to accomplish. And they don't have to know a whole lot about space technology, communications, you know, you know, you know, the differences between the [inaudible].

>> Maria Varmazis: Honestly, those kinds of capabilities do always seem like magic which seems cliche to me to even keep saying it. But it really just does because, same as you, I'm kind of like, how is this possible that this is real? But it is.

[ Music ]

>> Kathy O'Donnell: Yeah, so speaking about our machine learning and AI capabilities, so Earth's space is really interesting because they maintain a catalog over 10 million SAR satellite images. Now SAR is synthetic aperture radar. And SAR is pretty cool. So it can see through clouds. It can see and daytime, nighttime. It's not a traditional optical image like you would think of, but it still gives you a lot of detail.

Their catalog that they maintain on AWS, they're able to provide that and analyze that for government customers, first responders, any other user basically who can use that to create actionable insights. Again, it's, it's about the users creativity. Like you have all of this data. What can you do with it? And it comes down to just having that data available, being able to manipulate the compute, the analytics, build your machine learning models and immediately apply that to the data and get it out in front of people.

This actually ties in well to the demos that we're going to be doing at Reinvent. Reinvent is one of AWS's biggest events of the year for us and our customers. We're going to provide some demos for people interested in space, in satellites, and these demos actually allow them to play with these generative AI capabilities.

We're going to allow them to build lunar habitats, to build mission patches, to really experiment with the type of prompts that you can give to a diffusion model to create not just art, art's important, but things that you would start to refine to actually build. So you can say build me a lunar habitat, and you know, it will give you a pretty interesting picture. But then what really happens is you start refining what you're asking the model. Give me a lunar habitat that has these types of windows, these types of doors that would accommodate six to eight people. And then you really start drilling down, right.

>> Maria Varmazis: So the first prompt maybe would be the elimination of the dreaded blank page, right. Because that's always the hardest thing is, how do you get started? So then it, it gives you a starting point, but then you start to really refine. You go, I like that. I don't like that. Change this, change that. And you start to really --

>> Kathy O'Donnell: Exactly. That's exactly what happens. Yeah, so generative AI can get you kicked off. Start with some ideas. You can generate, like hundreds of ideas to go through. And then as you start seeing the things that make sense to you, that fulfill what your needs are, you can start putting those together and refining into a final version.

>> Maria Varmazis: That is fantastic, and that is often the hardest part of any major project is just that part. So making that easier and removing that friction is huge. And so I know also you're going to be doing a workshop on generative AI. So tell me a little bit about that.

>> Kathy O'Donnell: Yeah. So I'm pretty excited about this workshop. It's similar and yet different from the demos. So we are still using a generative AI foundation model. And what we're doing in this workshop is if you've ever been a coder, you've probably had to code out API specifications right, to turn, you know, what your user wants into a URL to pull from a server. It's not hard. It can take some time.

One of the issues is that it tends to be fragile. If the API specification changes, then you need to go in and you need to figure out what's happening. What we've seen though is that our generative AI large language models can actually take an API specification, take a natural language request from a user, and turn that into a URL to talk to the server with. A proper well formatted URL.

So in the workshop, we're going to show people how to change a user request like give me a picture of Jupiter from Tuesday into the actual URL to get that from the image server with just generative AI. So you're not- you're not hard coding any API specs in. You're not worried about making sure that this field is filled out correctly and that field, you let generative AI take care of it.

>> Maria Varmazis: That's incredible. So people attending this workshop, what will they take home with them or take back to work with them after doing that? What do you expect them to learn from that?

>> Kathy O'Donnell: Yeah, I think the main things they're going to learn, and this is something that has been a complete learning experience for me over the last few months working with generative AI is the power of prompt engineering. The way that you talk to generative AI is very different from the way that we talk to traditional machine learning models. The use of natural language, what it can figure out from context is very different. So from this, one of the big things we want people to take away is the different ways that you talk to generative AI. How important the prompts are that you send to a large language model. And then the. power of what that large language model can do based on those props.

>> Maria Varmazis: I have often heard the phrase prompt engineering and I I'm fascinated to hear you talking about that because it's ,it's one of those phrases, I don't know if everyone understands it. So to, to be able to give a better concrete understanding of what that really means is really valuable. Because there's a lot of noise around that right now.

>> Kathy O'Donnell: It is just so very different from the way that we've done things in the past. And using natural language to talk to a computer, to me, feels weird. And I start trying to give it, you know, much more computer type programming talk. And yeah, the large language models are fine with natural language. They say just talk to me like you would a person. I mean, they don't say that literally, but.

>> Maria Varmazis: I say do we use please and thank you when talking to a large language model?

>> Kathy O'Donnell: I believe some of them are built to be a little friendlier if you are friendly to them. So it really depends on the, on the model.

>> Maria Varmazis: So maybe, the answer is maybe. Alright.

>> Unidentified Speaker: I have this, I do tend to use please and thank you with the model. Like please summarize this document in less than 500 words, and then it does.

>> Unidentified Speaker: And then it does.

>> Maria Varmazis: Now we, we talked a lot about that, that incredible way that Earth and space can come together, and we can get these insights from space through the cloud with AI. I'm wondering what your vision is for how you could see this impacting people maybe in five years. Or what you hope people will take away from the, the insights that you've shared today?

>> Kathy O'Donnell: Yeah, well, I think, one of the main things I want to say is that we are like just at the beginning here of how space and cloud and generative AI can interact. And really what's exciting for me is that we let our customers lead the way. Like we let the people using AWS figure out for themselves, for their communities how to use this in the best and most creative ways possible. And so what's ahead in five years? We are going to make it so that anyone can use this technology. Because people know what's best for them. People are so creative. So I am just so excited to find out how people are going to use it. How are you going to bring space and cloud and our AI technology together to make life on Earth better, to make life in space better. So for me, that's what's really exciting about the next five years is just, I think, there is so much opportunity here, and I'm so excited to see where our customers go with it.

[ Music ]

And that's it for "AWS in Orbit" episode one. Kathy O'Donnell, AWS and generative AI. A special thanks to Kathy O'Donnell for joining us today. For additional resources from this episode, check out our show notes at space.n2k.com/aws. This episode was produced by Alice Carruth and powered by AWS. Our AWS producer is Laura Barber. Mixing by Elliott Peltzman and Tre Hester with original music and sound design by Elliot Peltzman. Our executive producer is Brandon Carpth [phonetic]. I'm Maria Varmazis. Tune in for a sneak preview of episode two, "Global Infrastructure at Scale" on November 14th. Thanks for listening.

[ Music ]